Classification Of Masses In Mammography Based On Information Fusion Of Multi-classifier And Multi-view

Breast cancer is one of the most frequently diagnosed malignant tumors of women in the world. In the latest twenty years, the incidence and mortality of breast cancer in China increase rapidly. And mammography is the most reliable method to detect breast cancer. Computer-aided diagnosis(CAD) system of the mammograms that can assist the radiologists as a second opinion to improve the detection efficiency has been studied extensively.Classification of masses in mammography is an important part of CAD. Improvements of accuracy and stability are the focus of the current classification studies. However, the performance of current CAD in mass detection remains relatively low and the radiologists could not have confidence in and accept this type of schemes. Because of the difference of masses in two views between mammography outline and background in clinical, classification models based on information fusion of multi-view and multi-classifier are studied in this dissertation to improve the classification accuracy, sensitivity, specificity, and stability. The main contents and innovations are as follows:(1) Multi-view fusion is used to classify masses in mammography. And experiments validate that it can improve the performance of classification.(2) Masses in mammograms are classified using multi-classifier fusion in single-view.(3) The outputs of each classifier in two views are fused, and then these results of multi-view are used for multi-classifier fusion.(4) Fusion of multi-classier is applied in each view, then the two fusion results are used for multi-view fusion.(5) Feature vectors of two views are fused, then classification of single-classifier and multi-classifier fusion are used.(6) According to the difference difficulty of distinguishing from benign and malignant of two views for the same mass, data set is devided into three groups. And evaluate classfication results of four models introduced in (2)—(5). The results indicate that the performance of (3) and (4) in terms of classification accuracy, sensitivity specificity and satbility is better than (2) and (5) because the former make full use of masses information in two views. Moreover, these two models are more in line with radiologist habits and can improve the clinical availability of CAD system. This study is based on the multi-view and multi-classifier fusion and lay a theoretical foundation on the CAD system for the realization of wide application in clinical.